4
$\begingroup$

I found this interesting Single RNA-seq data set in GEO, but I am not sure how to analyze it appropriately.

They have deposited transcriptomic profiles of human and mouse pancreatic islets (pancreatic cells: Beta cells, Delta, etc). The problem I see is that the different pancreatic cell types are not in equal numbers: 872 Beta cells and 214 Delta cells.

I want to see whether the expression of a set of genes are correlated between the two cell types. I don't think that I can visualize the correlation between the expression of the Beta and Delta cells on a scatter plot if they are unequal in number. Is there a workaround to include all the cells in the analysis, without losing good quality transcriptome of beta/delta cell?

What I would like to do is to compare the expression of two pancreatic cell types (Beta v.s. Delta cells) using a scatter plot. I have a set of genes that I am interested in, and would like to see their expression profile for beta and delta cells. As far as I know, it is only possible to visualize this using a scatterplot (X:Beta vs Y:Delta) where the number of isolated beta cells is the same as the number of delta cells.

Question: Given unequal number of isolated panacriatic cells, what would be an appropriate way to compare the expression profiles? Should I just ignore the extra ones?

Any idea how to approch this problem?

Thanks,

Improve this question
$\endgroup$
8
  • 2
    $\begingroup$ What is wrong with a traditional differential expression analysis? How big is the difference in number between Beta and Delta cells? $\endgroup$
    – llrs
    Dec 21, 2017 at 20:20
  • $\begingroup$ @Llopis Quite large ... the difference between beta and delta is 658 cells. $\endgroup$
    – MEhsan
    Dec 21, 2017 at 20:32
  • 1
    $\begingroup$ Do you have a particular set of genes in mind or are you hoping to go fishing for one? It's also unclear why you can't do visualizations with unequal sample sizes. $\endgroup$
    – Devon Ryan
    Dec 21, 2017 at 22:29
  • 2
    $\begingroup$ Any clarifications (such as what you have just written) should be put in the question, rather than the comment. $\endgroup$
    – gringer ♦
    Dec 21, 2017 at 23:06
  • 1
    $\begingroup$ @MEhsan You can randomly pick the same amount of delta cells than of beta cells and bootstrap to find the real correlation (or plot the scatter plot) $\endgroup$
    – llrs
    Dec 22, 2017 at 7:49

3 Answers 3

Reset to default
3
$\begingroup$

You are looking for a way to compare expression profiles between cell types, and your data is counts of genes expressed per cell. Your issue is that the data is highly dimensional. It has as many dimensions as you have genes. To be able find useful information, you must reduce these dimensions using a statistical technique such as Principal COmponents Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), or Multidimensional Scaling (MDS). The result of these analyses is a reduction of your thousands of dimensions to vectors (eigenvectors in PCA) of variance which you can plot against each other in a scatter plot.

I have been working with Single Cell RNA-seq data and have been exploring the difference in gene expression in a certain type of cells between timepoints. Seurat (R package) is a very powerful tool for scRNA-Seq data analysis.

Here is a a tSNE of my data. I have added the timepoint labels, but you can see that after performing the dimensionality reduction the two different timepoints cluster by their gene expression profiles. I can then use Seurat to see which specific genes are differentially expressed between the clusters and a whole bunch of other fun things.

tSNE plot of scRNA-Seq data

Improve this answer
$\endgroup$
1
1
$\begingroup$

Scatterplots are used to represent multiple dimensions of the same piece of data. A scatterplot would be used in this case to display all genes in your set on a single plot, but I would have some difficulty interpreting such a plot. If you want to look separately at each gene, then a density curve (or histogram) would be more appropriate.

Your problem is similar to common problems in flow cytometry, and I don't think there's any reason why similar approaches can't be used. Here's a histogram / density plot showing two different types of cells (negative control, and treatment condition), indicating the difference in fluorescence of a particular marker (somewhat analogous to gene expression) across all cells of each type. Note that the number of cells are different between the negative control and treatment:

Flow cytometry histogram

This image was taken from here, which also demonstrates a 2D scatter plot (where each point is a cell, rather than a gene).

Improve this answer
$\endgroup$
0
$\begingroup$

seurat3 may help. You can compare the marker genes from the 2 cell types or treat one cell type as a control. Please see the example analysis from: https://satijalab.org/seurat/v3.0/immune_alignment.html

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.